Datasets:

DOI:
License:
Dataset Viewer

The viewer is disabled because this dataset repo requires arbitrary Python code execution. Please consider removing the loading script and relying on automated data support (you can use convert_to_parquet from the datasets library). If this is not possible, please open a discussion for direct help.

Data for the SMeL Test benchmark (paper coming soon).

"qa" contains the synthetic documents used for the "ignoring dubious sources" and "resolving contradictions" tasks.

  • "ground_truth" contains seed documents from which all variants were generated
  • "rewritten" contains the rewritten source-specific variants
  • "corrupted" contains source-specific variants with alternative answers to corresponding questions

"summarization" contains the synthetic documents used for the "active filtering" task.

For usage examples, see our GitHub repo.

Downloads last month
6